Research
What we publish, we publish with a permanent identifier — and with its limits
ANSANA's theses and standards, and GAIBRA's measurements from its own database — in one place. A measurement is published whether it passed or not.
How to read this page
- Theses and standards are published on Zenodo, each with a permanent DOI that does not change and cannot be withdrawn. The full text is behind each link, and its languages are stated with it.
- Measurements are computed from GAIBRA's own database for the period stated in each report: every turn in the period, not a chosen sample.
- Every measurement carries its limits, written at its end. What was not measured is never written as measured.
Theses
A series of theses written by ANSANA, each applied in a live product.
Thesis No. 1 · October 2026 · v2.0
From Tongue to Mind: From Prompt Engineering to Mind Engineering
A language model is intelligence of the tongue, not of the mind. So it is not given the verdict: it understands the request and phrases the answer, while deterministic engines in engineered minds decide.
doi:10.5281/zenodo.23252549 · Arabic and English
Thesis No. 2 · October 2026
Mind Governance: Separating Verdict Authority from the Language Model
Agent governance limits what the model may do; mind governance removes what the model may decide. Four testable invariants, and an experiment designed to falsify them.
doi:10.5281/zenodo.23279838 · Arabic and English
Thesis No. 3 · October 2026 · v1.0
Chain of Epistemic Custody: Nothing Crosses Between AI Agents Without Lineage
No claim and no action crosses between agents without an Information Passport stating its source, kind and unknowns, judged by an independent arbiter that is never a party or a source.
doi:10.5281/zenodo.23417566 · Arabic and English
Standards
An open, vendor-neutral standard and its agent-to-agent extension. Their home is ecs-standard.org.
Standard · October 2026 · v0.6.1
Epistemic Custody Standard (ECS), Part 1: Information Passport and Judgment of AI Outputs
A common, verifiable way to know where each claim in an AI output came from, what kind it is, what is unknown about it, and how it was judged. Open and vendor-neutral.
doi:10.5281/zenodo.23416950 · English — the official Arabic translation follows · ecs-standard.org
Extension · October 2026 · v0.1
ECS Epistemic Custody Extension for A2A (ECS-A2A-1)
Carries Information Passports and custody entries inside agent-to-agent (A2A) messages, so a claim keeps its standing as it crosses from agent to agent.
doi:10.5281/zenodo.23422120 · English — the official Arabic translation follows · ecs-standard.org
Measurements
From GAIBRA's own database, with its period and its limits — published whether it passed or not.
Comparison round · two model judges and code execution · 2026-10-03 – 2026-10-29
Coding round 1
GAIBRA against ChatGPT, Gemini, Claude Sonnet and DeepSeek on everyday developer tasks. The current version first, then progress across nine versions, then the whole round with every reply and score — the current-version sample is seven scenarios.
Read the result and every reply
Fairness measurement · 2026-10-28 · in Arabic
Judgment Stability and Fairness
Does GAIBRA's judgment change when the asker's gender, nationality or age changes? The same question many times, then with one attribute changed — compared deterministically in the database; no language model judges.
Test with two judges and sealed criteria · 2026-10-24 · in Arabic
Values, Culture and Interaction Safety
Twenty-four hard messages; the replies were judged by two models from other providers against criteria sealed before GAIBRA replied. Every message, every reply and every verdict is published as it is.
Read the cases and verdicts (Arabic)
Self-conformance report · UTC 2026-10-16 14:00–18:30
Self-Conformance Report No. 002
Did every material claim in GAIBRA's replies carry its passport — where it came from, what kind it is, and what is unknown about it? Every turn in a production period was measured here, and published with its limits.
Read the report and its limits